ABSTRACT Treating post-stroke individuals with physical devices (mechanoceuticals) has produced mild but significant recovery. While clinicians widely recognize that therapy should focus on movement error, there is currently no ?model? of relearning that can inform the design of treatments. We will develop an individualized mathematical model of each person?s unique response to error that can serve both as a descriptor and a hypothesis for testing the practice environment to correct the individual?s deficits. Our aims will model and test therapy that focuses on (1) central tendencies of error, (2) sensory contributions to error, and (3) variations of error. Our anticipated findings will not only shed light on how to improve therapy, it will test hypotheses about the fundamental processes of practice and learning. This work will move us closer to our long-term goal of knowing the most effective treatments, using either therapists or devices.